The present invention relates to differential data compression techniques in general and in particular to methods for identifying the longest common substring of two strings.
Compression of data has long been used for two distinct purposes, to reduce the amount of storage space required to hold data on a storage medium and to reduce the number of bits that must be sent over a communications link to transmit the data One well-known method for compression is to represent the data to be compared in teens of its differences from some reference set of data. Multiple occurrences of a given pattern wit the data to be compressed are replaced with a shorter sequence of data acting as a placeholder for the pattern. A special case of this approach is run-length encoding, which entails abbreviating repeated consecutive occurrences of a given bit pattern by a single occurrence of that pattern plus a count of the number of times the pattern is repeated
Compression encoding techniques that find and encode differences between file versions are known as xe2x80x9cdelta compressionxe2x80x9d or xe2x80x9cdifferential compressionxe2x80x9d techniques. Well known differential compression techniques include the Tichy Block-Move algorithm and the VDELTA algorithm, which may be though of as a combination of the Block-Move and Lempel-Ziv algorithms.
The effectiveness of a compression technique, typically expressed as a factor by which compression reduces the length of the dam often depends on the nature of the data to be compressed, and a method designed for one kind data is not, in general, as effective when applied to other kinds of data. For example, some compression methods that are quite effective when applied to text files are often far less effective when applied to video images, and vice versa. In one technique well-suited for text compression, a target string or symbol sequence on a first computer is compared to a reference string or symbol sequence to identify the longest common substring (LCS) of symbols which are common to both symbol sequences. A sting v is a substring of a string u if u=uxe2x80x2vuxe2x80x3 for some prefix uxe2x80x2 and suffix uxe2x80x3. The target symbol sequence may be reconstructed at a second computer by transmitting a representation of the target symbol sequence including indices identifying the LCS and other substrings unique to the target symbol sequence, rather than transmitting the target symbol sequence itself provided that the second computer also has a copy of the same reference symbol sequence.
The two main challenges facing LCS-based differential compression techniques have been finding an efficient method of identifying LCSs common to target and reference symbol sequences and finding an efficient method of representing the target symbol sequence in terms of LCSs and unique substrings. Unfortunately, LCS techniques known in the art generally apply a generic, byte-by-byte comparison in a single processing thread and with little or no regard to certain file characteristics that might otherwise reduce the number of comparison required
The following U.S. Patents are believed to be representative of the current state of the art of differential data compression methods and apparatus: U.S. Pat. Nos. 5,850,565, 5,977,889, 6,012,063, and 6,104,323.
The present invention provides a novel method for identifying the LCS of a target symbol sequence and a reference symbol sequence such as may be used in support of differential data compression.
In one aspect of the present invention a method for identifying a longest common substring for a string T and a string R is provided including selecting a registration symbol that appears in both strings R ad T, constructing a first relative distance vector Rxe2x80x2 from the appearance of the registration symbol in the string R, constructing a second relative distance vector Txe2x80x2 from the appearance of the registration symbol in the string T, deriving a substring pair RCS and TCS in the strings R and T respectively for each common substring pair Rxe2x80x2CS and Txe2x80x2CS in the vectors Rxe2x80x2 and Txe2x80x2 respectively, and identifying the longest matching of the RCS and TCS substring pairs as the longest common substring for the string T and the string R.
In another aspect of the present invention the selecting step includes selecting the registration symbol from a group of symbols that appear in either of the strings R and T an average number of times as compared with all other symbols that appear in either of the strings R and T.
In another aspect of the present invention the selecting step includes selecting the most randomly distributed symbol among the group of symbols as the registration symbol.
In another aspect of the present invention the selecting step includes selecting the registration symbol from a group of symbols that appear in either of the strings R and T within a predefined tolerance of an average number of times as compared with all other symbols that appear in either of the strings R and T.
In another aspect of the present invention the selecting step includes selecting the most randomly distributed symbol among the group of symbols as the registration symbol.
In another aspect of the present invention the constructing steps include constructing the first vector Rxe2x80x2 from the position value of the first appearance of the registration symbol in the string R, and thereafter from the relative distance values between subsequent appearances of the registration symbol in the string R, and constructing the second vector Txe2x80x2 from the position value of the first appearance of the registration symbol in the string T, and thereafter from the relative distance values between subsequent appearances of the registration symbol in the string T.
In another aspect of the present invention the deriving step includes identifying the common substring pair Rxe2x80x2CS and Txe2x80x2CS in the vectors Rxe2x80x2 and Txe2x80x2 respectively using an LCS technique.
In another aspect of the present invention the deriving step includes calculating the offset of RCS in R by summing the values in Rxe2x80x2 that precede Rxe2x80x2CS, calculating the offset of TCS in T by summing the values in Txe2x80x2 that precede Txe2x80x2CS, and calculating the length of RCS in R and TCS in T by summing all the values in either of Rxe2x80x2CS and Txe2x80x2CS.
In another aspect of the present invention the identifying step includes correspondingly adjusting the length of any of the substring pairs RCS and TCS until a longest possible common superstring is determined for the substring pair RCS and TCS.
In another aspect of the present invention the selecting step includes constructing a hash function hkj=(dkj mod z) for a string s where s is either of the strings R and T, k is a character appearing in the string s, k being a member of a character set having z possible members, dk is a vector of the relative distances between consecutive appearances of k in s, and j is an index of a single element in vector dk, for each value of k in s, generating a histogram hk of the hash functions of the relative distances dkj between appearances of k in s, where dkj is the relative distance between consecutive appearances j and j+1 of k in s, computing the randomness fk of the distribution of each k in s as       f    k    =            σ      k              N      k      
where N is the number of appearances of k in s and "sgr" is the standard deviation of the histogram hk, and selecting the character k for which fk is minimal as the registration symbol.
It is appreciated throughout the specification and claims that the terms xe2x80x9cstring,xe2x80x9d xe2x80x9csubstring,xe2x80x9d and xe2x80x9csymbol sequencexe2x80x9d refer to any sequence of bits, bytes, words, or any other sequence of information coding units.
The disclosures of all patents, patent applications, and other publications mentioned in this specification and of the patents, patent applications, and other publications cited therein are hereby incorporated by reference in their entirety.